Sici?sici 0O14-3820%28 1205%3AWABEOT%3E2.0.C0%3B2-N

1206W. D. CRILL ET AL.tially have important, if rather diverse, effects on fitness. Forexample, body size and wing size often influence social dominance, mating success, fecundity, and flight dynamics (Robertson 1957; Partridge et al. 1987a,b; Starmer and Wolf1989). Abdominal melanization may influence heat absorption and potentially thermoregulatory capacity (Capy et al.1988). Knock-down temperature could index probability ofsurvival during a heat stress (Huey and Kingsolver 1993),and the thermal dependence of walking speed could affectsurvival or mating success (Christian and Tracy 1981; Partridge et al. 1987a). Finally, egg size can affect viability(Curtsinger 1976).Not surprisingly, all of the traits studied here were stronglyinfluenced by sex and developmental temperatures. However,a few traits were also influenced by maternal or paternaltemperatures. Recent companion studies from our lab demonstrate that both developmental and parental temperaturesinfluence male territorial success (Zamudio et al. 1995) andthat laying and paternal (but not developmental and maternal)temperature influenced fecundity of flies early in life (Hueyet al. 1995). This suite of experiments reinforces the viewthat adult phenotypes can sometimes be sensitive to betweenas well as within-generation environmental effects. Consequently, attempts to predict short-term evolutionary responses to selection are likely to be difficult (Gupta and Lewontin 1982; Riska et al. 1985; Kirkpatrick and Lande 1989).Moreover, attempts to measure "field heritabilities" (Coyneand Beecham 1987; Riska et al. 1989; Hoffmann 1991),which assume that cross-generational effects are insignificant, may be unreliable. Our observations of significant paternal effects should encourage studies of mechanism and ofthe adaptive significance (or lack thereof) of such cross-generational effects.The Flies Flies used in this experiment originated from a large population (- 1000 isofemale lines, courtesy of L. Harshman andM. Turelli) collected from Escalon, California in May 1991and maintained at the University of California, Davis at roomtemperature (-22 C) 13: 11 L:D cycle. We received a sample(-1000 flies) in April 1992. Thereafter, we reared flies invials at low and controlled density (-50 eggslvial; cornmeal,molasses, yeast, agar, tegosept) before transferring them topopulation cages of 2000 to 3000 individuals with discretetwo-week generations at 22 C on a 12: 12 L:D cycle. Presentexperiments were run in July through August 1992. The flieshad been in captivity for about 14 months and thus shouldhave been partially adapted to the laboratory environment(Service and Rose 1985).Overview of ProtocolWe independently manipulated paternal, maternal, and developmental temperatures of flies and then scored severaltraits of adults. We raised flies through two generations duringwhich we manipulated sire's temperature (Ts,,,; 18 C or2S C), dam's temperature (Tdam; 18 C or 25"C), and offspring's developmental temperature (Tdev;18OC or 25OC). Weselected 18 C and 25 C because these temperatures are wellwithin the natural range of developmental temperatures forD. melanogaster (Parsons 1978; McKenzie and McKechnie1979; Jones et al. 1987; Feder 1996) and these temperaturesresult in normal development (Economos and Lints 1986).We then crossed parental flies in the four possible sex-byparental temperature combinations and raised the eggs at either 18 C or 25 C. Thus, eight basic treatment groups (withdifferent thermal histories) were generated. The experimentwas replicated twofold, and 16 groups of flies were monitoredand scored for phenotypic traits.The Parental GenerationTo produce the parental generation, we collected eggs fromthe laboratory stock, transferred them into 40 vials (50 to 70eggstvial), and moved them to either 18 C or 25 C for development (Tsi,, and Tdam)Each egg-collection was limitedto five hours to reduce the exposure (hence possible acclimation) of eggs to 22 C before transfer to their respectivedevelopmental temperatures. Because development rate is inversely related to temperature, we staggered these initial eggcollections so that eggs raised at the two temperatures wouldeclose at similar times: consequently, we collected eggs destined for development at 18 C about ten days prior to collecting those destined for development at 25 C.After the parental flies eclosed, we collected virgin fliesduring the middle of the eclosion period, separated males andfemales (light C 0 2 anesthesia, 20 to 30 males or females pervial), and maintained them until ready for mating at theirrespective developmental temperatures (Tsireand Tdam).Because flies living at different temperatures will have different"physiological" ages (Long et al. 1980; Taylor 1981) andbecause parental age can influence offspring trait values (David 196% Parsons i964), we synchronized physiological agesof parental flies at mating. Thus 18 C parents were mated atfive days of adult age, whereas 25 C parents were mated atthree days of adult age (ages were scaled to relative development times at these two temperatures). Parental flies werecrossed at 22OC to minimize potential natural selection atdifferent temperatures and to ensure that all parental flies(and their eggs) experienced a small temperature shift. Werepeated the entire sequence to generate two replicate groupsof parents.The O m r i n g GenerationWe collected eggs for the offspring generation from theparental crosses (above) and transferred them to 18 C or 25 C(as above). The resulting flies belonged to eight replicatedexperimental groups (each with -2000 flies) that differed inpaternal, maternal, or developmental temperature (Table 1).Each of the groups of 18 C replicates eclosed after its paired25 C replicate (see below).We recognize two potential problems with our protocol.First, because these flies were outbred and were raised forone or two generations at different temperatures, they mighthave been exposed to inadvertent selection, such that differences among groups might reflect genetic as well as phenotypic effects. Such genetic differentiation should, however,be small. In pilot experiments, egg viability was invariably

1207BETWEEN- AND WITHIN-GENERATION EFFECTSTABLE1. Morphological measurements of adult flies as a function of parental and developmental temperatures. Values given are meansi standard errors. Sample sizes are shown in parentheses. For the various temperature treatments, H 25 C and L 18 C.TdevSexTdamTs, Dry mass(PSIhigh ( 90% of all eggs produced adults) at both 18 C and25"C, such that little viability selection could have occurredduring the egg stage. Moreover, the traits studied here haveonly moderate heritabilities (Roff and Mousseau 1987; Hueyet al. 1992). Nevertheless, some selection could have occurred either if genotypes differed in their mating success orfecundity as a function of temperature, or if mating successor fecundity was genetically correlated with traits scored here(Gupta and Lewontin 1982; A. A. Hoffmann, pers. comm.,1994). Second, we set up simultaneously all eggs for theoffspring generation (within a replicate) to ensure that alloffspring flies would start development at the same time andin the same batch of media. Consequently, however, 25 Cflies eclosed and were scored before the 18 C flies. Therefore,any observed differences between developmental-temperature groups might in part reflect measurement at differenttimes (Coyne et al. 1983). However, because observed differences between replicates were generally insignificant (seebelow), our conclusions are unlikely to be seriously confounded by temporal bias.Morphological and Life-History MeasurementsWing length(mm)Wing width(mm)Melanization(% black)Abdominal MelanizationWe scored the degree of melanization of the posterior portion of the abdomen (see David et al. 1990) for six femalesfrom each experimental group (C 96; all males have heavilymelanized abdomens). We used a camera lucida to trace thesilhouettes (lateral projected view) of the three posterior tergites and of the melanized portions, and then gravimetricallyestimated the percentage of the projected surface area thatwas melanized.Egg SizeFlies were initially maintained at their developmental temperature for three to five days posteclosion. They were thentransferred to 22 C for an additional 12 h, when we begancollecting freshly laid eggs. To facilitate handling, we partially submerged eggs in a drop of tap water and thus measured fully hydrated eggs. We recorded video images of 16to 20 eggs from each experimental group (C 3 14) and laterused a digitizing program to measure maximal length (1) andwidth (w) in arbitrary units of each egg.Physiological MeasurementsDry Body MassThermal Dependence of SpeedWe measured the dry mass of 10 flies of each sex randomlychosen from different vials, set up for each of the 16 experimental groups (C 320 flies). We dried flies at 60 C andthen weighed individual flies (those with complete wings andlegs) to the nearest kg.On the morning of testing, 15 males and 15 females fromeach replicated treatment group were selected haphazardlyand placed individually (without anesthesia) into plastic vials(70 X 15 mm). Speed was measured in a temperature-controlled, walk-in environmental chamber at the following sequence of temperatures 15", 25", lo0, 20 , 30 , 35 C. (Fliescan be permanently damaged by high temperatures, so performance at 35 C had to be measured last.) Vials were placedin the chamber 15 to 20 min prior to testing to ensure thatflies had equilibrated to the test temperature. Between trials,all individuals were held at 22 C.To measure walking speed, we knocked a fly down to thebottom of a vial, and recorded the elapsed time (later converted to velocity, cmls) until the fly reached to top of theWing SizeWe removed the right wing from 20 flies of each sex randomly selected from each experimental group (C 640). Wefixed the wings to glass slides (Partridge et al. 1987a), projected the images on a video monitor, and then used a digitizing program to measure the length (L3 vein) and the width(where L5 meets the wing margin) of each wing.

1208W. D. CRILL ET AL.vial. The resultant walking speeds are voluntary but appearto be near maximal performance levels. In effect, the fliesare exhibiting an escape response after being knocked offtheir perch. (This technique [Miquel 19761 also takes advantage of the negative geotropism of D. melanogaster.) Onlyone run per fly per temperature was used in about 95% ofthe tests. If, however, an individual flew rather than walked,up to two additional attempts were made. If the animal repeatedly flew, then the time taken for the last flight wasrecorded. At extreme temperatures some flies remained immobile for 60 sec (2% at 10 C; 14% at 3S C): these flieswere assigned a velocity of 0.0 cmls. All individuals wereanalyzed unless they failed to run at three or more adjacenttest temperatures.Knock-Down TemperatureWe measured the "knock-down" temperature, which is theupper temperature at which flies lose coordination and fallfrom a glass column (see Fig. 1 in Huey et al. 1992). Theapparatus used is a tall, water-jacketed glass column (withinternal baffles) connected to a heated water bath. The waterbath (and thus column temperature) was initially set to 30 C,which is a warm but not disabling temperature. We addedabout 1000 flies from a given group to the top of the columnand began pumping increasingly warm water through thesurrounding water jacket. The air inside the column heatedat a fairly constant rate (-0.7"CImin). Because Drosophilaare very small, their body temperature closely tracks the column's air temperature (see Appendix in Huey et al. 1992),which we monitored with a fine thermocouple.As the column heated, a fly became progressively warmerand eventually reached its "knock-down" temperature (Tkd)and then fell out the column into collecting tubes, which werechanged at 0.5"C intervals. We measured Tkd for a total of17,254 flies. Mean knock-down temperatures are generallyrepeatable (Huey et al. 1992; below).Because adult age affects heat resistance in Drosophila(Lamb and McDonald 1973), we standardized the physiological age of flies (as above). Specifically, we measuredknock-down temperatures on flies aged 5-6 days (18 C development) or 3-4 days (25 C development). All flies weretransferred to 22OC for 24 h prior to measurement so that allflies would experience the same acute temperature shift (22OCto 30 C) at the beginning of the knock-down experiment.teractions. With percent melanization and egg size, whichwere measured only for females, we initially used a fourfactor analysis of variance, with all two-way interactions. Ifthese initial analyses showed that replicates differed significantly, we included the replicate sum of squares in the errorsum of squares (Sokal and Rohlf 1981, p. 349). Abdominalmelanization was correlated with projected area of the tergites, so projected area was included as a covariate in themelanization analyses. Most analyses were performed usingS-Plus (StatSci Inc. 1993).In the measurements of knock-down temperature, about1000 flies all from the same treatment group were tested ina given run (above), such that data for individuals within runswere probably not fully independent. Therefore, we computedand subsequently analyzed only the mean knock-down temperature for males and for females from each run. This is anextremely conservative procedure because 17,254 measurements are reduced to only 32.The thermal dependence of sprint speed is a multivariatetrait, as performance is scored over multiple temperatures.We analyzed the data two ways. First, we analyzed threeperformance measures that summarize the position, height,and breadth of these performance curves (see Hertz et al.1983): To,, is the observed temperature at which an individualwalked fastest, umaxis the speed at Top,(thus maximal speed),and Tbris an index describing the breadth of the performancecurve. This latter index, which is derived from the secondmoment of area about a neutral axis, describes the distributionof performance about a central point, in this case about Top,:where ui is walking velocity at temperature Ti. (Velocities u,were standardized to umaxto remove spurious correlationsbetween Tbr and urnax.)We also used a repeated-measures ANOVA to analyze thermal performance curves. Intra-individual effects were nestedinside the effects for maternal, paternal, and developmentaltemperatures. This term was used as the error term for theamong-subjects analysis.Dry Body Mass and Wing SizeResultsStatistical AnalysesTo determine whether transformations were required, weinspected distributions of data and of residuals for normality.For dry body mass and knock-down temperature, no transformation was necessary. (In any case, we verified that standard transformations did not affect statistical conclusions.)Thermal performance breadths (see below) and maximalspeeds were log-transformed. Percent melanization data werearcsine (square root) transformed. For both wing and egg size,lengths and widths were correlated: we analyzed the principalcomponents (from z scores) for these two size measurements.With knock-down temperature and the size measurements,we initially ran a full factorial analysis of variance for Tsire,Tdam,Tdev,sex, and replicate with all possible two-way in-Dry body mass averaged 339 4.4 k g (x; SE. TableI), and the full model (Table 2) accounted for most of thevariation (79%) in dry mass. Sex had the dominant influenceon mass (P 0.0001): females were much larger than males(by 50%; Fig. 1). Developmental ( P 0.0007) and dam ( P 0.003) temperatures also influenced dry mass. Flies werealso relatively heavy if they developed at low temperatureor if their dam had been at low temperature (Fig. 1). However,these effects were small (e.g., development at 18 C increaseddry mass by only 4% relative to development at 25OC). Siretemperature had no significant effect ( P 0.17).The interaction between sex and developmental temperature was significant (Table 2, P 0.003): only female drybody mass was sensitive to developmental temperature. (A

1209BETWEEN- A N D W I T H I N - G E N E R A T I O N EFFECTSTABLE2.SourcedfA N O V A tables for morphological traits.Sum o f sqMean sqFvaluePr(F)Mass:SexT,,,,-ErrorW i n g size ( P C I ) :Sex1 re:TdevTdam:TdevErrorMelanization:A b d o m i n a l area-1 rate analysis only o f males verified that male size wasinsensitive to developmental temperature [ P 0.761.) Damb y developmental temperature was also significant ( P 0.0001): developmental temperature had a marked effectonlyi f the dam had been raised at 18OC but not at 25OC.Wing length and width were strongly correlated ( r 0.94),and so we analyzed only the first principal component score(97% o f variance) o f z-transformed measurements. The resulting A N O V A model (Table 2 ) accounted for 87% o f thevariation. Females had much larger wings than males ( P 0.0001, Table 2 ) . Developmental temperature also had a large,negative effect (Fig. 1, P 0.0001). However, neither dam( P 0.33) nor sire ( P 0.63) temperature influenced wingsize. None o f the two-way interactions was significant (all P 0.28, Table 2).DiscussionDevelopmental temperature had significant and inverse e f fects on dry body mass and on wing size o f females, but ononly wing size o f males. The insignificant response o f dry bodymass o f males may be idiosyncratic. In a separate study in ourlaboratory (Zamudio et al. 1995), but with a different stock o fD. melanogaster, male dry mass was significantly increased bydevelopment at 18 C versus 25 C. In any case, the inverse relationship between body size and developmental temperaturereiterates a classic pattern for Drosophila (David et al. 1983)and many other ectotherms (Atkinson 1994). (However, bodysize may be relatively small in flies that developed at extremelylow temperature [Economos and Lints 19841.)The mechanistic basis for the inverse relationship betweendevelopmental temperature and body size is o f interest (Partridge et al. 1995). Development at low temperature resultsin slow growth and in delayed maturation at a large size inmany organisms (Atkinson 1994). In contrast, developmenton low quality food also slows growth and delays maturation,but at a smaller-not larger-size (Gebhardt and Stearns1988). Several reaction-norm models for size and age at maturity as functions o f temperature or food quality have beenproposed (Stearns and Crandall 1984; Berrigan and Charnov1994). For example, Berrigan and Charnov (1994) analyzereaction norms in terms o f a general growth model, in whichlarger size (e.g., resulting from low temperature development) is assumed to result in increased fecundity. However,when females o f the stocks studied herein are raised at lowtemperature, they are relatively large (Table I ) , but nevertheless do not in fact have increased fecundity, at least earlyin life (Huey et al. 1995; below). Consequently, an evolutionary explanation for these reaction norms in D. melanogaster remains elusive.Evolutionary and developmental responses to temperatureare parallel. Drosophila evolving by laboratory natural selection at low temperatures become genetically larger than

W. D. CRILL ET AL.devdamsire**\PI*#*sex-101Wing size (PC1)Dry massdev***sire-101-101Egg size (PC1)MelanizationdevdamsuesexI-101Knockdown temperatureMaximum velocitydevdamsiresex-10iPerformance breadth-10iOptimum temperatureFIG. 1. Effects of treatments on eight phenotypic traits of Drosophila melanogaster. Solid bars represent the differences between theleast-squared means (scaled in standard deviation units) for treatment at 25 C minus that at 18 C (for sire, dam, and developmentaltemperature treatments). Unfilled bars represent the effect of sex (female minus male). Significance levels (from Tables 2, 4, and 6) areindicated adjacent to each bar (* 0.05, ** 0.01, *** 0.001).flies evolving at high temperature (Anderson 1966; Cavicchiet al. 1989; Partridge et al. 1995). Similarly, body size ispositively related to latitude (hence inversely related to meanambient temperature) in several species (David and Bocquet1975; Pegueroles et al. 1995).Wing size was more plastic with respect to developmentaltemperature than was dry mass (Fig. 1). Consequently, wingloading (mass per unit area of wing) may be reduced at lowdevelopmental temperatures. However, measurements of massand wing size on the same individuals are required to test this

1211BETWEEN- AND WITHIN-GENERATION EFFECTSexpectation. Wing loading is lower during cool seasons in D.melanogaster collected from nature (Stalker 1980) or in fliesraised at low temperature (Starmer and Wolf 1989). Severalpossible functional consequences are discussed in Starmer andWolf (1989; see also Curtsinger and Laurie-Ahlberg 1981).Cross-generational effects on body size in flies are generally thought to be small (Riska et al. 1989), but temperature-induced, cross-generational effects have not previouslybeen examined to our knowledge. In our studies, dry bodymass-but not wing size-was sensitive to dam temperature(Fig. 1). Flies whose mother had been raised at low temperature were slightly heavier, but had wings that were similarin length, than flies whose mother had been raised at lowtemperature (Fig. 1). (Note: a recent study confirms our finding that maternal temperature has no significant effect onwing length in D. melanogaster [A. A. Hoffmann, pers.comm., 19941.) In contrast, body mass and wing size wereinsensitive to sire temperature (Fig. 1). Whether the effectof dam temperature on body mass is adaptive requires investigation. The effect could reflect (in part) the larger eggsof low-temperature females (see below).Abdominal MelanizationResultsOn average, 42.6 2 1.27% of the posterior three tergitesof females were melanized in lateral projection view. Themodel accounted for 77% of the variation in melanization.Developmental temperature had a major impact: flies thatdeveloped at 18OC had 20% more of their projected surfacearea covered by melanin than did flies that developed at 25OC(Tables 1-2, P 0.0001, Fig. 1). None of the remaining mainfactors had a significant effect (all P 0.5, Table 2, Fig. 1).Of the two-way interactions, only the dam by developmentaltemperature was significant ( P 0.0009, Table 2): the effectof developmental temperature was accentuated if the damdeveloped at 18OC rather than at 25 C.DiscussionIncreased abdominal melanization in response to low developmental temperatures (Table 1) has been well documentedin Drosophila (David et al. 1985) and in other insects (Watt1990). Melanization also increases (genetically) with latitude

abdominal melanization, and reduced egg size. Parental temperatures influenced a few traits, but the effects were generally small relative to those of sex or developmental temperature. Flies whose mother had been raised at 25 C (versus at 18 C) had slightly higher knock-down temperature and smaller body mass. Flies whose father had been